Martin Krzywinski / Genome Sciences Center / Martin Krzywinski / Genome Sciences Center / - contact me Martin Krzywinski / Genome Sciences Center / on Twitter Martin Krzywinski / Genome Sciences Center / - Lumondo Photography Martin Krzywinski / Genome Sciences Center / - Pi Art Martin Krzywinski / Genome Sciences Center / - Hilbertonians - Creatures on the Hilbert Curve
Mad about you, orchestrally.Hooverphonicfeel the vibe, feel the terror, feel the painmore quotes

B1;2c UCD Computational and Molecular Biology Symposium, Dublin, Ireland. 1-2 Dec 2016.

art + science activism

Watch the video of this project, which features the participants who have a BRCA mutation and their interaction with the piece. The video also highlights the design and construction of the mural.

Martin Krzywinski @MKrzywinski
Science and art and personal stories of cancer survivors combine into this beautiful depiction of the complexity and individuality of the genome. (Free the Data)

Human Genome Art by Humans with Genomes

I recently took part in a deeply meaningful collaboration of science, art and personal stories of cancer survivors.

Together with Joanna Rudnick and Aaron De La Cruz, we sought to create a work of art that combines the science of cancer genomics and the individuals whose lives are affected by genetic mutations in the BRCA1 and BRCA2 genes, where genomic changes drastically increase one's chances of breast and ovarian cancer.

We wanted to make something that is scientifically accurate, artistically beautiful and emotionally engaging. The complexity of the genome, the multitudes of other genes and possible mutations and the millions of personal stories of hardship and survival were just a few of the elements we wanted to include the the piece.

My role was to provide the scientific direction behind the design and incorporate it into the aesthetic of Aaron De La Cruz, a street artist from San Francisco whose work echoes information, complexity, interaction and continuity. We all have a genome — a different genome. The ways in which our genomes are different is what gives us traits like hair and eye color, but is also what makes some of us predisposed to diseases like cancer.

The mural, which includes elements drawn by the cancer survivors, is part of the Free the Data campaign, which is advocating for an open access model of genome mutation databases so that scientists everywhere can analyze it and help women make informed choices about their breast-cancer risk.

The piece Importance of Data Sharing by Nature Methods illustrated the point:

Imagine you are a physician or researcher and seek to get more confirmation on the clinical impact of particular genetic variants. If your search of public databases comes up empty this does not necessarily mean that nothing is known about the mutations in question. Rather, the information may be locked away as a trade secret in a genetic testing company’s proprietary database.

The New York Times article DNA Project Aims to Make Public a Company’s Data on Cancer Genes captures the current state of the situation.

The mural was constructed on location at InVitae in San Francisco.

A video of the project is available.

Beautiful, meaningful and personal

This work will be, as far as I know, the first human annotation of mutations in the human genome by humans whose genomes have the mutations. That's quite a term!

I've always been mindful of the necessity of the mingling of art and science. In my work I tried to add things I felt about the science I thought to create work that combines our objective understanding of the world we live in with the subjective experience of living in it. This project, by far, has been the most keenly felt.

Martin Krzywinski @MKrzywinski
Adding emotion, keeping the science. (Free the Data)

the design

The mural was created in San Francisco on Saturday, July 13th, 2013. We are starting with a 11' x 6' wood canvas. These dimensions reflect the ratio of lengths of BRCA1 and BRCA2 proteins (1,863 and 3,418 amino acids, respectively)

Martin Krzywinski @MKrzywinski
The canvas aspect ratio reflects the ratio of BRCA1 and BRCA2 protein lengths. The proteins are represented on the canvas as lines. (Free the Data)

The BRCA1 and BRCA2 proteins are drawn on the canvas as straight-line sections.

Martin Krzywinski @MKrzywinski
The genes are depicted on the canvas as their protein products. (Free the Data)

The locations of the participants mutations are positioned on the protein lines as circles. For individuals with large deletions, the circle is placed at the first affected amino acid. Because BRCA1 is location on the opposite strand (anti-sense), its start on the canvas is on the right.

Martin Krzywinski @MKrzywinski
11 mutations, one for each of the cancer previvor and survivor participants, are placed on the protein lines as circles. The start of BRCA1 is on the right to reflect that this gene is on the anti-sense strand. (Free the Data)

The rest of the genome is now drawn. Aaron's style is perfect for depicting information and the endless complexity of the genome and its interacting elements. We were careful to include elements that indicate that the story told today is not complete. Millions of others have mutations in thousands of other genes, each potentially life-threatening. Just as the stories of our participants will continue to evolve, other stories are waiting to be told.

Martin Krzywinski @MKrzywinski
BRCA1 and BRCA2 proteins and their mutations, together with the rest of the genome. Other lines and circles hint at other genes, other mutations, as well as the biochemical interactions in the cells and personal interactions of those affected by the mutations. (Free the Data)

Once the "reference" genome is depicted, participants with BRCA1 and BRCA2 mutations will complete the art work by individually marking the positions of their mutations on the art using personalized colors. With Aaron's help, everyone created their own color by mixing primary colors.

Martin Krzywinski @MKrzywinski
Participants fill in their mutation circles with their personalized color. (Free the Data)

From base pair, to genome, to person, to life. All it takes is one tiny change in the genome to change a life forever.

Martin Krzywinski @MKrzywinski
The mutations of 11 people in the vastness of the genome. What's your story? (Free the Data)

creation of free the data mural

The BRCA1 and BRCA2 lines were placed on the canvas by first pinning two pieces of string, marked with the positions of the mutations.

Martin Krzywinski @MKrzywinski
String was used to mark the placing of lines and mutations. (Free the Data)

After drawing the protein lines, it was time to fill the canvas.

Martin Krzywinski @MKrzywinski
Aaron De La Cruz creating the art work. Here, he is filling the space in the canvas around the BRCA1 and BRCA2 segments with his design. The project was shot with a Red Camera—this is a sequence from its render application. (Free the Data)

Over the next 4 hours, Aaron filled in the canvas with the "rest" of the genome.

Martin Krzywinski @MKrzywinski
Aaron De La Cruz creating the art work. Here, he is filling the space in the canvas around the BRCA1 and BRCA2 segments with his design. The project was shot with a Red Camera—this is a sequence from its render application. (Free the Data)


Lucy, Karen, Steve, Ghecemy, Joanna, Jill, Lisa, Lynn, Ruth, Jenica, Susan

Cancer previvors and survivors who have been diagnosed with a mutation on BRCA1 or BRCA2 genes.

Joanna Rudnick (director/producer)

Joanna made her directorial debut with the Emmy-nominated In the Family, a deeply personal film about coming to terms with testing positive for the breast cancer gene BRCA1 mutation and following the storylines of other women and families facing the same hard choices. In the Family premiered at Silverdocs in 2008, was broadcast nationally on PBS P.O.V. the same year and was a finalist for the NIHCM Foundation’s Health Care Radio and Television Journalism Award.

Joanna received a master’s degree in Science and Environmental Journalism from New York University and a bachelor’s degree in English from Northwestern University. Joanna loves the opportunity to teach and mentor and served as an adjunct professor at Northwestern University’s Medill School of Journalism in the past.

She has written for several publications including Audubon Magazine, The Artful Mind, The Berkshire Record and Humanities. Before finding her way to the wonderful world of documentaries, Joanna served as an Americorps volunteer, implementing project-based environmental curricula in the San Francisco Public School System.

Joanna is one of the cancer survivors whose mutations are encoded in the art.

Aaron De La Cruz (artist)

Aaron De La Cruz's work, though minimal and direct at first, tends to overcome barriers of separation and freely steps in and out of the realms of design, graffiti, and illustration.

The parameters he has chosen to work within actually allow him to free himself and react to the very limitations he has created. This overriding structure and the lack of deliberation while moving within creates a tension when encountering his work due to the almost computer generated grid like systems he creates by unplanned markmaking. The act and the marks themselves are very primal in nature but tend to take on distinct and sometimes higher meanings in the broad range of mediums and contexts they appear in and on.

Martin Krzywinski @MKrzywinski
Work by Aaron De La Cruz. (Aaron De La Cruz)

His work finds strengths in the reduction of his interests in life to minimal information. De La Cruz gains from the idea of exclusion, just because you don't literally see it doesn't mean that its not there.


VizUm: Colin Ware and Martin Krzywinski — register for free.


news + thoughts

Model Selection and Overfitting

Tue 13-09-2016

With four parameters I can fit an elephant and with five I can make him wiggle his trunk. —John von Neumann.

By increasing the complexity of a model, it is easy to make it fit to data perfectly. Does this mean that the model is perfectly suitable? No.

When a model has a relatively large number of parameters, it is likely to be influenced by the noise in the data, which varies across observations, as much as any underlying trend, which remains the same. Such a model is overfitted—it matches training data well but does not generalize to new observations.

Martin Krzywinski @MKrzywinski
Nature Methods Points of Significance column: Model Selection and Overfitting (read)

We discuss the use of training, validation and testing data sets and how they can be used, with methods such as cross-validation, to avoid overfitting.

Altman, N. & Krzywinski, M. (2016) Points of Significance: Model Selection and Overfitting. Nature Methods 13:703-704.

Background reading

Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Classifier evaluation. Nature Methods 13:603-604.

Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Logistic regression. Nature Methods 13:541-542.

...more about the Points of Significance column

Classifier Evaluation

Tue 13-09-2016

It is important to understand both what a classification metric expresses and what it hides.

We examine various metrics use to assess the performance of a classifier. We show that a single metric is insufficient to capture performance—for any metric, a variety of scenarios yield the same value.

Martin Krzywinski @MKrzywinski
Nature Methods Points of Significance column: Classifier Evaluation (read)

We also discuss ROC and AUC curves and how their interpretation changes based on class balance.

Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Classifier evaluation. Nature Methods 13:603-604.

Background reading

Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Logistic regression. Nature Methods 13:541-542.

...more about the Points of Significance column

Happy 2016 `\pi` Approximation, roughly speaking

Sun 24-07-2016

Today is the day and it's hardly an approximation. In fact, `22/7` is 20% more accurate of a representation of `\pi` than `3.14`!

Time to celebrate, graphically. This year I do so with perfect packing of circles that embody the approximation.

By warping the circle by 8% along one axis, we can create a shape whose ratio of circumference to diameter, taken as twice the average radius, is 22/7.

If you prefer something more accurate, check out art from previous `\pi` days: 2013 `\pi` Day and 2014 `\pi` Day, 2015 `\pi` Day, and 2016 `\pi` Day.

Logistic Regression

Tue 13-09-2016

Regression can be used on categorical responses to estimate probabilities and to classify.

The next column in our series on regression deals with how to classify categorical data.

We show how linear regression can be used for classification and demonstrate that it can be unreliable in the presence of outliers. Using a logistic regression, which fits a linear model to the log odds ratio, improves robustness.

Martin Krzywinski @MKrzywinski
Nature Methods Points of Significance column: Logistic regression? (read)

Logistic regression is solved numerically and in most cases, the maximum-likelihood estimates are unique and optimal. However, when the classes are perfectly separable, the numerical approach fails because there is an infinite number of solutions.

Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Logistic regression. Nature Methods 13:541-542.

Background reading

Altman, N. & Krzywinski, M. (2016) Points of Significance: Regression diagnostics? Nature Methods 13:385-386.

Altman, N. & Krzywinski, M. (2015) Points of Significance: Multiple Linear Regression Nature Methods 12:1103-1104.

Altman, N. & Krzywinski, M. (2015) Points of significance: Simple Linear Regression Nature Methods 12:999-1000.

...more about the Points of Significance column

Visualizing Clonal Evolution in Cancer

Thu 02-06-2016

Genomic instability is one of the defining characteristics of cancer and within a tumor, which is an ever-evolving population of cells, there are many genomes. Mutations accumulate and propagate to create subpopulations and these groups of cells, called clones, may respond differently to treatment.

It is now possible to sequence individual cells within a tumor to create a profile of genomes. This profile changes with time, both in the kinds of mutation that are found and in their proportion in the overall population.

Martin Krzywinski @MKrzywinski
Ways to present temporal and phylogenetic evolution of clones in cancer. M Krzywinski (2016) Molecular Cell 62:652-656. (read)

Clone evolution diagrams visualize these data. These diagrams can be qualitative, showing only trends, or quantitative, showing temporal and population changes to scale. In this Molecular Cell forum article I provide guidelines for drawing these diagrams, focusing with how to use color and navigational elements, such as grids, to clarify the relationships between clones.

Martin Krzywinski @MKrzywinski
How to draw clone evolution diagrams better. M Krzywinski (2016) Molecular Cell xxx:xxx-xxx. (read)

I'd like to thank Maia Smith and Cydney Nielsen for assistance in preparing some of the figures in the paper.

Krzywinski, M. (2016) Visualizing Clonal Evolution in Cancer. Mol Cell 62:652-656.